diff options
| -rw-r--r-- | doc/neps/new-iterator-ufunc.rst | 18 |
1 files changed, 9 insertions, 9 deletions
diff --git a/doc/neps/new-iterator-ufunc.rst b/doc/neps/new-iterator-ufunc.rst index 76b21d644..6c4bb6488 100644 --- a/doc/neps/new-iterator-ufunc.rst +++ b/doc/neps/new-iterator-ufunc.rst @@ -29,7 +29,7 @@ Key benefits include: A large fraction of this iterator design has already been implemented with promising results. Construction overhead is slightly greater (a.flat: -0.5 us, newiter(a): 1.4 us and broadcast(a,b): 1.4 us, newiter([a,b]): +0.5 us, nditer(a): 1.4 us and broadcast(a,b): 1.4 us, nditer([a,b]): 2.2 us), but, as shown in an example, it is already possible to improve on the performance of the built-in NumPy mechanisms in pure Python code together with the iterator. One example rewrites np.add, getting a @@ -596,7 +596,7 @@ branch, where libndarray has the array named ``NpyArray`` and functions named ``NpyArray_*``. The iterator is named ``NpyIter`` and functions are named ``NpyIter_*``. -The Python exposure has the iterator named ``np.newiter``. One possible +The Python exposure has the iterator named ``np.nditer``. One possible release strategy for this iterator would be to release a 1.X (1.6?) version with the iterator added, but not used by the NumPy code. Then, 2.0 can be release with it fully integrated. If this strategy is chosen, the @@ -1670,7 +1670,7 @@ First, here is the definition of the ``luf`` function.:: nargs = len(args) op = args + (kwargs.get('out',None),) - it = np.newiter(op, ['buffered','no_inner_iteration'], + it = np.nditer(op, ['buffered','no_inner_iteration'], [['readonly','nbo_aligned']]*nargs + [['writeonly','allocate','no_broadcast']], order=kwargs.get('order','K'), @@ -1719,7 +1719,7 @@ Python iterator protocol.:: def iter_add_py(x, y, out=None): addop = np.add - it = np.newiter([x,y,out], [], + it = np.nditer([x,y,out], [], [['readonly'],['readonly'],['writeonly','allocate']]) for (a, b, c) in it: @@ -1732,7 +1732,7 @@ Here is the same function, but following the C-style pattern.:: def iter_add(x, y, out=None): addop = np.add - it = np.newiter([x,y,out], [], + it = np.nditer([x,y,out], [], [['readonly'],['readonly'],['writeonly','allocate']]) while not it.finished: @@ -1770,7 +1770,7 @@ of the iterator, designed to help speed up the inner loops, is the flag def iter_add_noinner(x, y, out=None): addop = np.add - it = np.newiter([x,y,out], ['no_inner_iteration'], + it = np.nditer([x,y,out], ['no_inner_iteration'], [['readonly'],['readonly'],['writeonly','allocate']]) for (a, b, c) in it: @@ -1807,7 +1807,7 @@ using other NumPy machinery to efficiently execute it. Let's modify ``iter_add`` once again.:: def iter_add_itview(x, y, out=None): - it = np.newiter([x,y,out], [], + it = np.nditer([x,y,out], [], [['readonly'],['readonly'],['writeonly','allocate']]) (a, b, c) = it.itviews @@ -1890,7 +1890,7 @@ Here's the same function, rewritten to use a new iterator. Note how easy it was to add an optional output parameter.:: In [5]: def composite_over_it(im1, im2, out=None, buffersize=4096): - ....: it = np.newiter([im1, im1[:,:,-1], im2, out], + ....: it = np.nditer([im1, im1[:,:,-1], im2, out], ....: ['buffered','no_inner_iteration'], ....: [['readonly']]*3+[['writeonly','allocate']], ....: op_axes=[None,[0,1,np.newaxis],None,None], @@ -1965,7 +1965,7 @@ implement the composite operation with numexpr.:: This beats the straight NumPy operation, but isn't very good. Switching to the iterator version of numexpr, we get a big improvement over the -straight Python function using the iterator. Note that in this is on +straight Python function using the iterator. Note that this is on a dual core machine.:: In [29]: def composite_over_ne_it(im1, im2, out=None): |